Routine, systematic measurement of food waste in acute care hospitals is challenging, particularly when monitoring plate waste at scale. This study examines the accuracy of a digital photography method for measuring aggregate solid-food plate waste across hospital mealtimes, with the aim of determining its suitability as a practical, scalable tool for ongoing monitoring.
Plate waste was assessed across seven breakfasts (n = 108), 21 lunches (n = 804) and 21 dinners (n = 851), reflecting the full hospital menu cycle (one-week rotation for breakfast and three-week rotation for lunch and dinner). For each meal, waste was measured using both direct weighing and visual estimation based on a 7-point scale applied to pre- and post-consumption photographs. Waste quantities for individual solid-food components were calculated by multiplying visually estimated leftovers (%) by standard serving sizes. Agreement between methods was assessed using Bland-Altman plots and Spearman correlation.
Digital photography estimates showed a strong positive correlation with direct weighing across all mealtimes (rs = 0.9895, p < 0.001), with consistently high agreement at breakfast, lunch and dinner (all p < 0.001). Overall, the mean difference between methods was 4.67 g, with limits of agreement from −50.51 to 59.84 g. Breakfast showed the highest accuracy, while lunch and dinner showed slightly wider limits. Findings support digital photography as a valid and efficient alternative to direct weighing for monitoring solid-food plate waste, with greatest accuracy at breakfast.
Digital photography offers practical advantages by reducing labour, time and spatial requirements, supporting its feasibility and sustainability for routine plate waste monitoring.
Abbreviation
1. Introduction
Food waste is an inherent issue within the hospital sector, accounting for up to 50% of total waste generated in some healthcare facilities (Cook et al., 2023). Plate waste is a major component of food waste in hospitals and is estimated to be two to three times higher than in other foodservice sectors (Cook et al., 2022). In this context, plate waste refers to the served food that is discarded and not consumed by patients (Williams and Walton, 2011). The causes of plate waste are many and varied including clinical factors (e.g. poor appetite, special diets, dysphagia), food and menu-related issues (e.g. over-sized portions, poor food quality and presentation, unfavourable temperature, unpleasant texture, limited food choice), service-related challenges (e.g. packaging difficulties, ordering errors, negative staff interactions, lack of feeding assistance, incorrect items delivered), and environmental considerations (e.g. poorly timed meals, distracting ward conditions, insufficient time to eat, and meal interruptions) (Antasouras et al., 2023). Measuring plate waste in a hospital setting is crucial to evaluating patient's nutritional intake and informing menu design (Heighington-Wansbrough and Gemming, 2022; Martins et al., 2014). Elevated levels of plate waste are associated with reduced energy and protein intake, as well as lower intakes of other essential nutrients, contributing to increased risks of weight-loss, malnutrition and other adverse clinical outcomes (Schuetz et al., 2021; Barton et al., 2000). Plate waste assessments provide valuable data to government and hospital administrators, offering guidance on foodservice management, menu planning and responsive interventions. These assessments also highlight opportunities to reduce food costs and improve operational efficiency (Connors and Rozell, 2004). Beyond clinical and economic implications, the environmental burden of producing food that remains unconsumed is considerable; thus, reducing plate waste not only reduces the volume of waste sent to landfill but also mitigates greenhouse gas emissions, particularly methane emissions from decomposing food, and reduces the loss of natural resources including land, water and energy used throughout the food supply chain (Seberini, 2020).
Plate waste in hospitals has traditionally been measured using direct weighing and visual estimation methods. With direct weighing, each plate or tray can be weighed before and after patient consumption, with the difference in weight recorded as individual plate waste. At the hospital level, total unconsumed food from all patients can also be weighed to estimate aggregate waste (Williams and Walton, 2011). While direct weighing is considered the most accurate method, assuming proper calibration and use of scales, it is labour-intensive, time-consuming and requires dedicated space for implementation (Food Loss and Waste Protocol, 2016). These practical constraints become particularly more challenging in large-scale or high-throughput settings such as hospitals (Giboreau et al., 2019).
Furthermore, complexity increases when multiple food components (e.g. main dishes, starch, vegetables) are served on the same plate (Pouyet et al., 2015). Upon return, leftovers are typically mixed, making it impossible to separate and weigh individual food components accurately (Pouyet et al., 2015). Thus, quantifying plate waste at the item-level is often impractical using direct weighing alone, particularly in busy hospital settings (Pouyet et al., 2015). Alternative methods such as visual estimation have been used to overcome such limitations of direct weighing (Martin et al., 2014).
The visual estimation method involves the use of a visual scale to measure the proportion of food left on the plate (Williams and Walton, 2011). This method is faster, less labour-intensive, and space-efficient than direct weighing, and can be performed through direct observation or digital photography (Giboreau et al., 2019; Williamson et al., 2004; Byker Shanks et al., 2017). The direct observation approach requires trained assessors to be present in the kitchen or dining area to observe patients' meal trays, and estimate the proportion of food waste left on the plate in real time (Ngo et al., 2009). The digital photography method entails photographing plates or trays before and after consumption, with visual estimates made later based on assessment of these images (Pouyet et al., 2015).
The digital photography method offers several advantages over the direct visual estimation when measuring plate waste (Simmons and Reuben, 2000). Direct visual estimation is often rushed and less accurate due to workflow constraints and time pressures in hospital foodservice settings, where trays must be cleaned rapidly (Parent et al., 2012). In contrast, digital photography enables efficient data collection with minimal disruption to foodservice staff (Simmons and Reuben, 2000). This method provides a durable visual record of meals before and after consumption without the need to rush the estimation of food leftover on the plate (Parent et al., 2012). It also allows for plate waste to be rated by multiple assessors, supporting reliability checks (Williamson et al., 2003; Giboreau et al., 2019; Simmons and Reuben, 2000). In addition, assessors do not need to recall tray contents prior to delivery to wards, eliminating memory-related inaccuracies (Simmons and Reuben, 2000). Overall, the digital photography method appears better suited to hospital environments, balancing accuracy with practicality when compared to direct visual estimation.
While various methodologies are available for quantifying plate waste, a recent review highlighted considerable variability in measurement approaches and emphasised the urgent need for validated, standardised methods to facilitate meaningful comparisons of food waste across diverse operational contexts to ensure reproducibility in future interventions (Amicarelli and Bux, 2021). Additionally, evidence indicates that routine measurement itself can promote waste reduction behaviour, further underscoring the importance of adopting practical and valid methods within institutional foodservice environments (Ramos et al., 2024).
Only a limited number of studies have validated the digital photography method against the direct weighing method in hospitals (Monacelli et al., 2017; Winzer et al., 2018) or nursing homes (Pouyet et al., 2015), and just one has measured aggregate plate waste specifically (Pouyet et al., 2015). Prior research in healthcare settings has typically assessed the method's accuracy for measuring food or nutritional intake at select mealtimes (e.g. lunch only), against the direct weighing method. To date, no study has validated the digital photography method for measuring plate waste.
This study examines the accuracy of a digital photography method for measuring aggregate solid-food plate waste in an acute care hospital across mealtimes, with the aim of determining its suitability as a practical, scalable tool for ongoing plate waste monitoring. A 7-point visual scale was used to estimate food waste by comparing pre- and post-consumption photographs of patient trays. The first objective was to compare plate waste estimates derived from the digital photography method with those obtained using direct weighing. The second objective was to assess whether the accuracy of the digital photography method varied across mealtimes (breakfast, lunch and dinner).
2. Materials and methods
2.1 Study design and setting
This study was conducted in an acute care hospital in Adelaide, South Australia, between September and October of 2023. This study consisted of two distinct phases: the first phase implemented a quality assurance protocol through a portion size assessment to establish baseline portion accuracy; the second phase validated the digital photography method against the direct weighing method for whole plate waste assessment.
At the time of data collection, inpatient food production relied on a cook-fresh system. Meals were organised according to a three-week rotational menu for lunch and dinner and a one-week rotation for breakfast. Patients selected their meals from printed menus provided by Nutrition and Dietetic Assistants. Foodservice staff received all essential plating information, including chef summaries and tray labels, prior to meal assembly.
Serving sizes were standardised in accordance with the Menu and Nutrition Standards for Public Health Facilities in South Australia (SA Health, 2020). Kitchen facilities were equipped with calibrated portioning tools and utensil guides tailored to specific food components, such as “wet main dish”, “single main item”, “mashed sweet potato and mashed potato”, “vegetables”, “texture-modified vegetables”. These measures aimed to support consistency in meal presentation and portion adherence.
2.2 Meal characteristics
For breakfast, patients could select up to three food components from a set menu comprising scrambled eggs, grilled tomato and either baked beans or spaghetti. Baked beans and spaghetti were served on alternating days within a seven-day breakfast cycle. Grilled tomato was unavailable on weekends, during which patients were limited to choosing from two components.
For lunch and dinner, patients were able to select one item from each of the following categories: main dish (e.g. “Apricot Chicken”, “Battered Fish”, “Lamb Curry”), starch (e.g. “Steamed Rice”, “Mashed Potato”, “Rosemary Potato”), green or brassica vegetables (e.g. “Broccoli”, “Cauliflower”, “Green Beans”), orange or red vegetables (e.g. “Pumpkin”, “Carrot”, “Sweet Corn”) or mixed vegetables available as an alternate to individual vegetable options. Patients were allowed to choose one of each from all food categories at each meal. This structured categorisation facilitated consistent portioning and alignment with hospital nutrition standards.
2.3 Quality assurance protocol for pre-consumption portion size validation
The digital photography method assumed minimal variation in portion sizes, relying on standardised plating procedures. To validate this, a quality assurance protocol was implemented. This included validation of commonly served menu items through direct weighing of randomly selected plated components prior to meal service across five days, which were then compared against hospital-standard serving sizes. Five to ten samples per food component were assessed, with samples collected from each meal (breakfast, lunch, dinner) to balance representativeness with practical limitations. Each food item was weighed with a calibrated digital scale, and the actual weights were compared against menu standards to assess consistency and reliability. Detailed quality assurance procedures for validating pre-consumption portion sizes are provided in the Supplementary Material Appendix 1.
2.4 Assessment of plate waste
Plate waste was measured in two general medicine wards and one coronary care unit using both direct weighing and digital photography methods. These wards were selected in consultation with hospital dietetics and foodservice staff due to their higher proportion of patients capable of consuming meals, the availability of diverse diet types and a broader menu spread compared with other wards, providing greater opportunity to capture variability in patient meal ordering and plate waste. Data were collected across one week for breakfast (aligned with the one-week rotation menu) and three consecutive weeks for lunch and dinner (aligned with the three-week rotation menu). The direct weighing method was used as the standard reference (Food Loss and Waste Protocol, 2016).
2.4.1 Direct weighing method
The direct weighing method steps included: (1) Labelling each meal tray with a unique identifier before plating to ensure traceability upon return of the tray after the meal; (2) Retrieving returned trays directly from meal trolleys; (3) Weighing leftover food using an electronic scale; (4) Calculating the weight of the leftover food by subtracting the weight of an empty plate (determined from the average of 20 identical plates) from the total weight of the plate with leftover food.
2.4.2 Digital photography method
A 7-point visual scale was used to assess plate waste based on pre- and post-consumption photographs. The scale was selected to balance estimation resolution with practicality, allowing meaningful differentiation in the proportion of food left, while remaining feasible for application across a large sample. Compared with coarser scales (3-, 4-, 5- and 6-point formats), the additional intermediate categories may improve sensitivity to partial consumption, without increasing assessor burden. When applied at the food component level, this approach enabled consistent, robust and replicable estimation of aggregate plate waste.
The digital photography method included these steps: (1) Labelling each meal tray with unique identifiers before plating; (2) Photographs of each meal tray were taken pre-delivery and post-return using a smartphone (Android, VIVO IQOO Neo8) under routine foodservice conditions within an enclosed hospital kitchen using standard artificial lighting. In the fast-paced plating environment, a mobile device was selected over fixed camera or tripod systems to avoid disruptions to the routine workflow, particularly during the plating process, where a high volume of meals are prepared. Pilot testing demonstrated that the use of a fixed camera set-up slowed tray delivery and interfered with normal operations, whereas mobile phone photography enabled rapid image capture without requiring specialised equipment. Tripods were not used due to space constraints along the plating line, where multiple foodservice staff simultaneously plate individual food components and a distribution supervisor verifies tray accuracy; (3) To ensure consistency, all photographs were taken from directly above the tray surface at a broadly standardised distance. All meals were served on uniform round white plates with a rim and placed on consistent light-grey meal trays. (4) Paired before- and after-consumption photographs were compared (Figure 1), to estimate the percentage of leftover food for each item (excluding sauces) using a 7-point visual scale: 0% (none left), 10% (one mouthful left), 25% (onequarter left), 50% (one-half left), 75% (three-quarters left), 90% (one mouthful eaten) and 100% (all food left) (Sherwin et al., 1998); (5) Standard serving sizes for each food component (main dish, starch, green/brassica vegetables, orange/red vegetables and mixed vegetables) were obtained from the Menu and Nutrition Standards for Public Health Facilities in South Australia (SA Health, 2020); (6) Plate waste quantity for each food component was calculated by multiplying the visually estimated percentage (%) of food left on the plate by the corresponding standard serving size for each food component; (7) Aggregate plate waste per tray was calculated by summing up estimated waste across all food components. Table A.1. in the Supplementary material (Appendix 1) demonstrates the calculation process for visually estimating plate waste from the breakfast photograph example and its comparison with the corresponding direct weighing result.
The six-image panel is arranged in two rows and three columns. The top row (“A 1”, “B 1”, “C 1”) shows three plated hospital meals as served, and the bottom row (“A 2”, “B 2”, “C 2”) shows the corresponding plates after eating, illustrating the amount and distribution of plate waste. In “A 1,” the served meal consists of scrambled eggs, spaghetti in tomato-based sauce, and a grilled tomato half. In “A 2,” the same plate after eating shows partially consumed spaghetti, eggs, and tomato. In “B 1,” the served meal includes a Irish lamb stew, broccoli florets, and diced potato bake. In “B 2,” the plate after eating shows scattered small pieces of meat and potato, with a small portion of broccoli remaining and sauce residues visible on the plate. In “C 1,” the served meal contains rice, sliced carrots, butter beans, and a diced pork dish in peanut sauce. In “C 2,” the plate after eating shows small amounts of rice, vegetables, and sauce remaining, with most of the meal consumed.Examples of pre-consumption and post-consumption plate photographs for breakfast, lunch and dinner. (A1) Pre-consumption breakfast plate photograph; (A2) Post-consumption breakfast plate photograph; (B1) Pre-consumption lunch plate photograph; (B2) Post-consumption lunch plate photograph; (C1) Pre-consumption dinner plate photograph; (C2) Post-consumption dinner plate photograph. Source(s): Authors' own work
The six-image panel is arranged in two rows and three columns. The top row (“A 1”, “B 1”, “C 1”) shows three plated hospital meals as served, and the bottom row (“A 2”, “B 2”, “C 2”) shows the corresponding plates after eating, illustrating the amount and distribution of plate waste. In “A 1,” the served meal consists of scrambled eggs, spaghetti in tomato-based sauce, and a grilled tomato half. In “A 2,” the same plate after eating shows partially consumed spaghetti, eggs, and tomato. In “B 1,” the served meal includes a Irish lamb stew, broccoli florets, and diced potato bake. In “B 2,” the plate after eating shows scattered small pieces of meat and potato, with a small portion of broccoli remaining and sauce residues visible on the plate. In “C 1,” the served meal contains rice, sliced carrots, butter beans, and a diced pork dish in peanut sauce. In “C 2,” the plate after eating shows small amounts of rice, vegetables, and sauce remaining, with most of the meal consumed.Examples of pre-consumption and post-consumption plate photographs for breakfast, lunch and dinner. (A1) Pre-consumption breakfast plate photograph; (A2) Post-consumption breakfast plate photograph; (B1) Pre-consumption lunch plate photograph; (B2) Post-consumption lunch plate photograph; (C1) Pre-consumption dinner plate photograph; (C2) Post-consumption dinner plate photograph. Source(s): Authors' own work
2.4.3 Inter-rater reliability assessment
A total of 1763 plates were assessed using the digital photography method by the primary assessor, distributed across breakfast (n = 108), lunch (n = 804) and dinner (n = 851). To assess inter-rater reliability, 200 plates were randomly selected: breakfast (n = 20), lunch (n = 90) and dinner (n = 90). Each plate included two photographs (pre- and post-consumption), resulting in 400 photographs for cross-assessment by the primary assessor and an additional independent assessor.
The selected trays encompassed a variety of main dishes, starches, and vegetables reflecting the full variety offered on the hospital menu. Both assessors were qualified in food science and nutrition and had experience in foodservice operations and food handling. The primary assessor was a trained nutrition and foodservice researcher, while the independent assessor also held a Master of Dietetics and was an Accredited Practicing Dietitian and Nutritionist. At the time of study, both assessors were PhD candidates with prior experience in nutrition research.
Prior to formal data collection, the primary assessor completed calibration exercises involving 50 practice plates covering breakfast, lunch and dinner. During this process, the accuracy of visual estimates for aggregate plate waste was assessed by direct comparison with weighed measurements. The independent assessor completed training using 10 practice plates across breakfast, lunch and dinner meals. Training involved immediate feedback based on pre-and post-consumption meal tray photographs and comparison with the primary assessor's estimates. Inter-rater consistency exceeding 80% agreement was required before formal estimation commenced. Both assessors independently rated the photographs using the same 7-point visual scale consistently throughout the study, while blinded to both the direct weighing results and each other's estimates. A statistical assessment of the consistency of the two assessors' visual estimates was then conducted.
2.5 Statistical analysis
Statistical analyses were performed using SPSS (version 29.0.1; IBM Corp., Armonk, NY, USA) and GraphPad Prism 10. A Chi-Square test was used to examine the relationship between plate composition and meal type (breakfast, lunch and dinner). Descriptive statistics, including mean, median and interquartile range (IQR) were calculated for aggregate (whole) plate waste. The mean and standard deviation of actual portion sizes were calculated for each food component.
To assess the consistency of actual food portions with hospital menu standards, the percentage deviation was calculated for each assessed food component using the formula:
Positive percentages indicate over-portioning, and negative percentages indicate under-portioning. A margin of error within ±10% was considered acceptable based on the hospital quality assurance thresholds (Paquet et al., 2003). Therefore, the compliance rate was calculated as the proportion of assessed samples per food component falling within ±10% of the hospital menu standards.
To assess the measurement discrepancy between methods, the percentage difference between direct weighing and digital photography was calculated using the formula:
A Mann–Whitney U test was performed to compare differences in plate waste estimates between the two methods. Inter-rater reliability between the two assessors was calculated based on estimates for individual food components and aggregate plate waste. Specifically, the Intra-class Correlation Coefficient (ICC) was used by applying a two-way mixed effects model, single measure, absolute agreement, with 95% confidence interval (Pouyet et al., 2015). ICC values were interpreted according to Koo and Li (2016): poor reliability: <0.5, moderate reliability: 0.5–0.75, good reliability: 0.75–0.90, and excellent reliability: >0.90 (Koo and Li, 2016).
The Spearman rank correlation coefficient was also computed to examine the correlation between plate waste measured using direct weighing and digital photography methods. A p-value less than 0.05 was considered statistically significant.
Validation of the digital photography method against direct weighing was conducted using the Bland-Altman approach, which assesses the average difference between two quantitative measures and estimates the interval within which 95% of the differences are expected to lie (Giavarina, 2015). For each paired observation, the difference between methods was plotted against the means from digital photography and direct weighing measurements. The 95% limits of agreement were calculated as:
Agreement between methods was evaluated on the basis of bias (mean difference between two methods), precision (standard deviation of the differences) and limits of agreement (mean difference ± 1.96 standard deviation of the differences). If approximately 95% of observations fell within the limits of agreement, the two measurement methods are considered interchangeable (Bland and Altman, 2010).
3. Results
3.1 Number of meals analysed and meal characteristics
Over the study period, a total of 108 breakfast plates, 808 lunch plates and 856 dinner plates were delivered to the two general medicine wards and one coronary care unit. Nine meal images (four lunch and five dinner) were excluded due to missing before or after meal photographs, preventing accurate matching. In total, 3,526 images from 1763 plates were analysed by the primary assessor, consisting of 216 breakfast images, 1,608 lunch images and 1702 dinner images. Meals included four texture categories: general texture, easy to chew, minced and moist, and soft and bite-sized. The majority were general texture (79%), followed by easy to chew (13%). The plate composition characteristics for the meals are presented in Table 1. This varied with some meals containing a single item while others comprised multiple food components. Plate composition differed significantly by mealtime, Of the plates analysed (n = 1763), 41% contained four food items and 28% contained three food items – predominantly during lunch and dinner. In contrast, 21% contained two food items and 10% contained one item, more commonly associated with breakfast (Table 1).
Characteristics of plate composition across breakfast, lunch and dinner
| Breakfast plates n (%) | Lunch plates n (%) | Dinner plates n (%) | All plates n (%) | |
|---|---|---|---|---|
| One component | 68 (63%) | 53 (7%) | 55 (6%) | 176 (10%) |
| Two components | 33 (31%) | 177 (22%) | 169 (20%) | 379 (21%) |
| Three components | 7 (6%) | 277 (34%) | 203 (24%) | 487 (28%) |
| Four components | 0 (0%) | 297 (37%) | 424 (50%) | 721 (41%) |
| Total | 108 | 804 | 851 | 1763 |
Note(s): Values are presented as number of plates (n, %). The association between meal type and plate composition was examined using the Chi-square test
3.2 Quality assurance of pre-consumption portion sizes
The actual portions of 153 food samples were assessed against menu standards across breakfast (n = 40), lunch (n = 55) and dinner (n = 58) over a five-day period. Breakfast components assessed included baked beans, spaghetti, scrambled eggs and grilled tomato. For lunch and dinner, the assessment covered three categories of meat dishes (solid piece, wet dish without vegetables and wet dish mixed with vegetables), starch, green or brassica vegetables, orange or red vegetables, and mixed vegetables.
Table 2 shows the portion accuracy for each food component. The mean deviation ranged from −11.0% to +6.9% across all food components, with only scrambled egg exceeding the acceptable ±10% threshold, indicating slight under-portioning (Table 2). The overall compliance rate was 82.4%, with the highest observed during dinner (87.9%), followed by lunch (81.8%) and breakfast (75%). Variation in compliance was observed across food components, with baked beans and a solid piece of meat having the highest compliance (100%), while scrambled egg showed the lowest (40%) (Table 2).
Summary of portion size audit results for different food components
| Food component | Number of samples (n) | Menu standard (g) | Mean ± SD actual weight (g) | Mean deviation∗ (%) | Compliance rate∗∗ (%) |
|---|---|---|---|---|---|
| Baked beans | 10 | 130 | 135.9 ± 8.1 | +4.5 | 100 |
| Spaghetti | 10 | 130 | 139.0 ± 5.4 | +6.9 | 80 |
| Scrambled eggs | 10 | 130 | 115.7 ± 3.1 | −11.0 | 40 |
| Grilled tomato | 10 | 70 | 66.8 ± 5.2 | −4.6 | 80 |
| Solid piece of meat (meat, poultry or fish) | 10 | 100 | 98.3 ± 5.4 | −1.7 | 100 |
| Wet dish-meat and sauce (no vegetables) | 15 | 130 | 129.4 ± 7.1 | −0.5 | 93.3 |
| Wet dish (mix of meat and vegetables) | 18 | 150 | 142.7 ± 6.7 | −4.9 | 94.4 |
| Starch | 20 | 90 | 85.8 ± 7.0 | −4.7 | 75 |
| Green/brassica vegetables | 20 | 60 | 61.2 ± 4.6 | +1.9 | 85 |
| Orange/red vegetables | 20 | 60 | 60.8 ± 6.8 | +1.3 | 70 |
| Mixed vegetables | 10 | 120 | 119.7 ± 9.1 | −0.3 | 90 |
Note(s): *Mean deviation (%) represents the percentage difference between the actual food portion and the hospital menu standard
**Compliance rate (%) indicates the proportion of servings within ±10% of the hospital menu standard
3.3 Comparison of plate waste between digital photography and direct weighing methods
Plate waste measurements derived from both methods are presented in Table 3. Across all meals, the mean plate waste ranged from 60.8 g/plate to 146.4 g/plate for the direct weighing method and 66.0 g/plate to 140.8 g/plate for the digital photography method across meals (Table 3). The mean difference in plate waste between direct weighing and digital photography methods ranged from −5.2 g/plate to 8.7 g/plate (Table 3).
Comparison of plate waste measured using direct weighing and digital photography methods by meals (breakfast, lunch, dinner)
| Mealtime | Direct weighing (g/patient/day) | Digital photography (g/patient/day) | Difference (g)* | % Difference* | p-value** | ||||
|---|---|---|---|---|---|---|---|---|---|
| Mean | Median (IQR) | Mean | Median (IQR) | Mean | Median (IQR) | Mean | Median (IQR) | ||
| Breakfast | 60.8 | 28.0 (0.0–106.0) | 66.0 | 26.3 (0.0–101.9) | −5.2 | 0.0 (−6.0–0.0) | −4.0 | 0.0 (−6.9–0.0) | 0.854 |
| Lunch | 139.3 | 114.0 (0.0–245.8) | 134.2 | 107.5 (0.0–240.0) | 8.7 | 0.0 (0.0–11.0) | 2.4 | 0.0 (0.0–6.4) | 0.666 |
| Dinner | 146.4 | 121.0 (0.0–256.0) | 140.8 | 120.0 (0.0–250.0) | 7.6 | 0.0 (0.0–10.0) | 1.9 | 0.0 (0.0–5.7) | 0.625 |
| Overall | 137.9 | 108.0 (0.0–242.0) | 133.2 | 104.0 (0.0–240.0) | 5.0 | 0.0 (0.0–7.0) | 1.4 | 0.0 (0.0–4.5) | 0.587 |
Note(s): *The negative sign indicates that the digital photography method overestimated breakfast plate waste; The positive sign indicates that the digital photography method underestimated lunch plate waste
**p values were calculated using the non-parametric Mann–Whitney U test. p-values <0.05 were considered statistically significant
Both methods showed that less plate waste was generated at breakfast, more at lunch, and the most at dinner. Statistical analysis showed no significant differences in plate waste between the two methods overall or by meal type (p > 0.05).
3.4 Correlation between plate waste measured using digital photography and direct weighing
Figure 2 shows the correlations between plate waste estimates derived from the digital photography and direct weighing methods across all meal types. A strong positive correlation was observed overall (rs = 0.9895, p < 0.001) as well as for each individual meal type: breakfast (rs = 0.9950, p < 0.001), lunch (rs = 0.9882, p < 0.001) and dinner (rs = 0.9882, p < 0.001). These results illustrate high agreement between the two measurement techniques across different meal types.
The four scatter plots are arranged in a two-by-two grid labeled “Breakfast”, “Lunch”, “Dinner”, and “Overall”. In each plot, the horizontal axis is labeled “Plate waste by direct weighing method (gram)” and the vertical axis is labeled “Plate waste by photography method (gram)”. Each plot also reports a Spearman correlation coefficient (“r s”) with statistical significance (“P less than 0.001”). In the “Breakfast” plot, both axes range from 0 to 400 grams with 100-gram increments. The reported correlation is “r s equals 0.9950 (P less than 0.001)”. The scatter points form a clear upward diagonal pattern, indicating a strong positive relationship between the two measurement methods. In the “Lunch” plot, both axes range from 0 to 600 grams with 100-gram increments. The correlation is “r s equals 0.9882 (P less than 0.001)”. The points are densely distributed but follow a strong upward trend, with slightly greater horizontal dispersion at higher plate waste values. In the “Dinner” plot, both axes range from 0 to 600 grams with 100-gram increments. The correlation is “r s equals 0.9882 (P less than 0.001)”. The scatter points show a similarly strong positive linear pattern with moderate horizontal spread at larger plate waste amounts. In the “Overall” plot, both axes range from 0 to 600 grams with 100-gram increments. The correlation coefficient is “r s equals 0.9895 (P less than 0.001)”. The points are tightly clustered along an upward-sloping line, indicating strong agreement between the two methods across all meals combined, with less variability than seen in the individual meal plots.Correlations between plate waste measured using the digital photography method and the direct weighing method across different meals (breakfast, lunch, and dinner). Associations were assessed using Spearman's rank correlation coefficient. Each data point represents an observation. Source(s): Authors' own work
The four scatter plots are arranged in a two-by-two grid labeled “Breakfast”, “Lunch”, “Dinner”, and “Overall”. In each plot, the horizontal axis is labeled “Plate waste by direct weighing method (gram)” and the vertical axis is labeled “Plate waste by photography method (gram)”. Each plot also reports a Spearman correlation coefficient (“r s”) with statistical significance (“P less than 0.001”). In the “Breakfast” plot, both axes range from 0 to 400 grams with 100-gram increments. The reported correlation is “r s equals 0.9950 (P less than 0.001)”. The scatter points form a clear upward diagonal pattern, indicating a strong positive relationship between the two measurement methods. In the “Lunch” plot, both axes range from 0 to 600 grams with 100-gram increments. The correlation is “r s equals 0.9882 (P less than 0.001)”. The points are densely distributed but follow a strong upward trend, with slightly greater horizontal dispersion at higher plate waste values. In the “Dinner” plot, both axes range from 0 to 600 grams with 100-gram increments. The correlation is “r s equals 0.9882 (P less than 0.001)”. The scatter points show a similarly strong positive linear pattern with moderate horizontal spread at larger plate waste amounts. In the “Overall” plot, both axes range from 0 to 600 grams with 100-gram increments. The correlation coefficient is “r s equals 0.9895 (P less than 0.001)”. The points are tightly clustered along an upward-sloping line, indicating strong agreement between the two methods across all meals combined, with less variability than seen in the individual meal plots.Correlations between plate waste measured using the digital photography method and the direct weighing method across different meals (breakfast, lunch, and dinner). Associations were assessed using Spearman's rank correlation coefficient. Each data point represents an observation. Source(s): Authors' own work
3.5 Inter-rater reliability of plate waste estimates
Inter-rater reliability was assessed for plate waste estimates provided by two independent assessors across breakfast, lunch and dinner meals. Reliability was evaluated for both total plate waste and individual food components using intraclass correlation coefficients (ICC) (Table 4). ICC values ranged from 0.835 to 0.992, indicating good to excellent agreement. The highest reliability for overall plate waste was observed at dinner (ICC = 0.972), while lunch showed the lowest (ICC = 0.874). For individual components, dishes with sauces (spaghetti/baked beans) had the lowest level of agreement (ICC = 0.835). Whole foods such as grilled tomato had the highest level of agreement (ICC = 0.992).
Intra-class correlation coefficient (ICC) of estimates of total plate waste and plate waste by individual food components between the two assessors
| Mealtimes | Food components | ICC (95%CI)∗ | p-value |
|---|---|---|---|
| Breakfast | All items | 0.924 (0.822–0.969) | <0.001 |
| Scrambled egg | 0.938 (0.665–0.982) | <0.001 | |
| Grilled tomato | 0.992 (0.980–0.997) | <0.001 | |
| Hot dishes (spaghetti or baked beans) | 0.835 (0.516–0.952) | <0.001 | |
| Lunch | All items | 0.874 (0.815–0.916) | <0.001 |
| Main dish | 0.953 (0.929–0.969) | <0.001 | |
| Starch | 0.914 (0.871–0.943) | <0.001 | |
| Vegetables | 0.912 (0.869–0.941) | <0.001 | |
| Dinner | All items | 0.972 (0.958–0.982) | <0.001 |
| Main dish | 0.979 (0.968–0.986) | <0.001 | |
| Starch | 0.963 (0.944–0.976) | <0.001 | |
| Vegetables | 0.960 (0.939–0.974) | <0.001 |
Note(s): ∗ ICC values were calculated using a two-way mixed effects model for absolute agreement. Values were interpreted as: <0.5 poor, 0.5–0.75 moderate, 0.75–0.90 good, >0.90 excellent reliability (Koo and Li, 2016)
3.6 Bland-Altman agreement analysis between digital photography and direct weighing methods
Bland-Altman plots illustrate agreement between plate waste estimates derived from the digital photography and the direct weighing across meal types (Figure 3).
The four Bland–Altman plots are arranged in a two-by-two grid labeled “Breakfast”, “Lunch”, “Dinner”, and “Overall”. In each plot, the horizontal axis is labeled “Average plate waste by both methods (gram)” and the vertical axis is labeled “Difference in plate waste (gram)”. A horizontal red line indicates the mean difference (bias). The upper horizontal green line indicates the upper limit of agreement, calculated as the mean difference plus 1.96 standard deviations. The lower horizontal green line indicates the lower limit of agreement, calculated as the mean difference minus 1.96 standard deviations. In the “Breakfast” plot, the horizontal axis ranges from 0 to 400 grams in 100-gram increments, and the vertical axis ranges from negative 100 to positive 100 grams in 50-gram increments. The mean difference is negative 5.22 grams. The upper limit of agreement is 16.62 grams and the lower limit of agreement is negative 33.96 grams. The data points are moderately scattered around zero. In the “Lunch” plot, the horizontal axis ranges from 0 to 600 grams with 200-gram increments, and the vertical axis ranges from negative 200 to positive 200 grams with 100-gram increments. The mean difference is 5.05 grams. The upper limit of agreement is 61.09 grams and the lower limit of agreement is negative 50.99 grams. The points are densely clustered around the zero-difference line, with slightly wider spread at higher average plate waste values. In the “Dinner” plot, the axes use the same ranges and increments as the Lunch plot. The mean difference is 5.56 grams, with a upper limit of agreement of 62.02 grams and a lower limit of agreement of negative 50.91 grams. The points show a similar distribution, with moderate variability around the mean difference. In the “Overall” plot, the horizontal axis also ranges from 0 to 600 grams, and the vertical axis ranges from negative 200 to positive 200 grams in 100-gram increments. The mean difference is 4.67 grams. The upper and lower limits of agreement are 59.84 grams and negative 50.51 grams, respectively. The points are densely clustered near the zero-difference line, indicating close agreement between the photography and weighing methods overall, with slightly greater variability at higher plate waste values.Bland-Altman plots illustrating the agreement between the direct weighing method and the digital photography method in measuring plate waste across meal types (breakfast, lunch, and dinner). Red lines represent the mean difference between direct weighing and digital photography methods. Green lines represent the 95% limits of agreement (Mean ± 1.96 SD). Source(s): Authors' own work
The four Bland–Altman plots are arranged in a two-by-two grid labeled “Breakfast”, “Lunch”, “Dinner”, and “Overall”. In each plot, the horizontal axis is labeled “Average plate waste by both methods (gram)” and the vertical axis is labeled “Difference in plate waste (gram)”. A horizontal red line indicates the mean difference (bias). The upper horizontal green line indicates the upper limit of agreement, calculated as the mean difference plus 1.96 standard deviations. The lower horizontal green line indicates the lower limit of agreement, calculated as the mean difference minus 1.96 standard deviations. In the “Breakfast” plot, the horizontal axis ranges from 0 to 400 grams in 100-gram increments, and the vertical axis ranges from negative 100 to positive 100 grams in 50-gram increments. The mean difference is negative 5.22 grams. The upper limit of agreement is 16.62 grams and the lower limit of agreement is negative 33.96 grams. The data points are moderately scattered around zero. In the “Lunch” plot, the horizontal axis ranges from 0 to 600 grams with 200-gram increments, and the vertical axis ranges from negative 200 to positive 200 grams with 100-gram increments. The mean difference is 5.05 grams. The upper limit of agreement is 61.09 grams and the lower limit of agreement is negative 50.99 grams. The points are densely clustered around the zero-difference line, with slightly wider spread at higher average plate waste values. In the “Dinner” plot, the axes use the same ranges and increments as the Lunch plot. The mean difference is 5.56 grams, with a upper limit of agreement of 62.02 grams and a lower limit of agreement of negative 50.91 grams. The points show a similar distribution, with moderate variability around the mean difference. In the “Overall” plot, the horizontal axis also ranges from 0 to 600 grams, and the vertical axis ranges from negative 200 to positive 200 grams in 100-gram increments. The mean difference is 4.67 grams. The upper and lower limits of agreement are 59.84 grams and negative 50.51 grams, respectively. The points are densely clustered near the zero-difference line, indicating close agreement between the photography and weighing methods overall, with slightly greater variability at higher plate waste values.Bland-Altman plots illustrating the agreement between the direct weighing method and the digital photography method in measuring plate waste across meal types (breakfast, lunch, and dinner). Red lines represent the mean difference between direct weighing and digital photography methods. Green lines represent the 95% limits of agreement (Mean ± 1.96 SD). Source(s): Authors' own work
Overall, the mean difference between methods was 4.67 g (SD = 28.15), with 95% limits of agreement ranging from −50.51 g to 59.84 g (Figure 3). The mean bias between methods was small and consistent across meals (breakfast: −5.22 g; lunch: 5.05 g; dinner: 5.56 g). However, limits of agreement varied by meal type, with breakfast showing narrower limits of agreement (−33.96 to 23.51 g) compared to lunch (−50.99 to 61.09 g) and dinner (−50.91 to 62.02 g), suggesting lower variability and greater accuracy in breakfast estimates. When lunch and dinner were combined, the mean difference was 5.31 g (SD = 28.70), with limits of agreement ranging from −50.93 g to 61.55 g. Separate analysis of lunch and dinner yielded similar results, indicating no substantial difference in agreement between lunch and dinner.
Overall, the mean difference between the two methods corresponded to a mean of 3.4% plate waste measured by the direct weighing. The Bland-Altman plots showed strong clustering around the equality line, with only 7% of plates exceeding 95% limits of agreement (2% reflecting overestimation and 5% underestimation). This was consistent across lunch and dinner, with breakfast plates showing slightly higher outliers, with 8% of plates exceeding the limits (7% reflecting overestimation and 1% underestimation).
4. Discussion
This study compared digital photography and direct weighing for measuring aggregate solid-food plate waste in an Australian acute care hospital. No significant differences were found between plate waste measurements across meal types, with minimal discrepancies overall. These findings support digital photography as an accurate, efficient and reliable approach to measuring aggregate solid-food plate waste in hospital settings.
Food-component level visual estimation was adopted over whole-plate estimation, informed by prior research showing stronger agreement between food-component estimates and direct weighing (Kawasaki et al., 2016). Studies in long-term care facilities similarly highlight limitations of whole plate estimation and advocate for more nuanced techniques (Castellanos and Andrews, 2002; Pokrywka et al., 1997). Incorporating food-component level estimation thus strengthened the validity of the digital photography method.
Although a strong correlation between methods suggests linear association, it does not confirm agreement (Giavarina, 2015). Bland-Altman analysis revealed a small mean difference (4.67 g), with 93% of data points within limits of agreement, indicating satisfactory concordance. This approach offers a nuanced understanding of measurement validity by visualising central tendency and patterns of difference (Giavarina, 2015). Inter-rater reliability was assessed using ICC, with values ranging from 0.835 to 0.992, indicating good to excellent reliability and strong replicability in future applications (Koo and Li, 2016).
To our knowledge, no prior study has validated digital photography against direct weighing for aggregate plate waste across all hospital mealtimes. Existing research focused on food intake at selected meals or individual items. Three studies (Winzer et al., 2018; Monacelli et al., 2017; Pouyet et al., 2015) compared the two methods at lunch only, using individual food items. This study's inclusion of breakfast, lunch and dinner offers a novel contribution.
Pouyet et al. (2015) assessed four lunch dishes using three assessors and mixed visual scales. Our study used a single primary assessor across the full-menu cycle with a 7-point scale, reducing the inter-assessor variation. A second assessor scored a random sample of 200 plates to evaluate reliability. In addition, Pouyet et al. (2015) relied on a single pre-consumption reference photo for all plates, whereas we captured both pre- and post-consumption images for each plate, helping to minimise plating variability. Other studies (Winzer et al., 2018; Monacelli et al., 2017), focused on individual items (e.g. soups, mains), with accuracy assessed for a single food component rather than whole plate waste. This approach is less applicable in hospital settings where meals are served as mixed plates (Williams and Walton, 2011). After consumption, components are often intermixed, complicating isolation and estimation. Single-item analysis also overlooks the complexity of mixed meals.
Digital photography was more accurate for breakfast than lunch or dinner, likely due to simpler meal composition. In our study, 94% of breakfast plates had one or two food components, compared to 71 and 74% of lunch and dinner plates, which typically included three or four components, respectively. Breakfast meals were also more standardised in presentation, which may have facilitated clearer visual comparison between pre- and post-consumption images. In contrast, lunch and dinner meals involved greater dish complexity, including multiple components and mixed dishes, which likely increased the difficulty of visual estimation. This increased complexity may have contributed to greater variability and reduced accuracy for lunch and dinner compared with breakfast. These findings highlight the importance of considering meal type when interpreting plate waste estimates and when applying the digital photography method for routine plate waste monitoring.
Food component-level estimation is more difficult with intermixed or overlapping items. Variations in colour, shape and arrangement can bias visual assessment. For example, the Delboeuf illusion theory suggests light-coloured foods on white plates may be underestimated, while vivid colours attract disproportionate attention (Rabiei and Nazari, 2023). Uniform food shapes (e.g. spherical mashed potato) yield less error than irregular ones (e.g. chicken curry) (Parent et al., 2012). Overlapping food components, where one item partially conceals another, may further increase the difficulty of accurate estimation. To improve accuracy, benchmark photographs of plates with known quantities of waste (via direct weighing) could be incorporated in future studies. Although divided plates may enhance the precision of estimating plate waste, they do not resolve challenges posed by mixed dishes and may negatively affect the dining experience.
Capturing both pre-meal and post-meal photographs also increases assessor workload and risks data loss if image pairs are mismatched (Hinton et al., 2013; Pouyet et al., 2015). In our study, mismatches accounted for less than 1% and were excluded. Despite compositional challenges, digital photography remains more convenient, time-efficient and resource-friendly than direct weighing (Heighington-Wansbrough and Gemming, 2022). It enables identification of high-waste food components, supports menu optimisation and improves patient satisfaction through better alignment of meals with patient preferences. Clinically, routine monitoring of plate waste may also help identify patients at risk of malnutrition for further nutritional assessment and intervention (Pouyet et al., 2015; Winzer et al., 2018). AI-powered visual estimation (e.g. camera bin technology) may address current limitations, such as delays and subjectivity and empower patients to reflect on their intake and dietary habits (Liu et al., 2025). However, this system requires significant upfront investment, staff training, and ongoing maintenance. These factors may limit its practicality and affordability for many hospital foodservices at this stage. Beyond these operational and clinical benefits, the feasibility of the digital photography method in estimating plate waste also has important implications for sustainability in hospital foodservice systems.
4.1 Implications for sustainability and hospital foodservice management
Although this study focuses on methodological evaluation, food waste in hospital foodservices has important implications for environmental sustainability and resource efficiency (Chatzipavlou et al., 2024). Food waste generated during the consumption phase (plate waste) often reflects avoidable resource use (water, land, energy, labour, chemicals and materials), and contributes to greenhouse gas emissions from disposal, thereby exacerbating climate change (Tonini et al., 2018). Empirical evidence from hospital settings demonstrates the magnitude of these impacts. For example, a recent study conducted in Iran estimated that each kilogram of plate waste generated in hospitals was associated with approximately 8.1 m2 of land use, 1.4 kg of CO2-equivalent greenhouse gas emissions and 1,003 L of water wastage (Anari et al., 2024). Similarly, hospital-based assessment reported that plate waste generated per hospital bed per day resulted in emissions of 0.85 kg CO2-equivalent, the loss of 580.25 L of fresh water, and 3.08 g of nitrogen released to the environment (Abiad et al., 2025). These findings highlight that plate waste is not only a nutritional or operational issue but also a significant contributor to the environmental and resource burden of hospital foodservice systems.
Given the magnitude of environmental impacts associated with hospital plate waste, accurate, feasible and affordable measurement is an essential prerequisite for developing effective food waste reduction strategies. However, previous studies have consistently reported that direct weighing of plate waste, while accurate, is often too time-consuming for routine use in busy hospital environments, where capturing waste from each item separately can be impractical (Ni et al., 2026; Williams and Walton, 2011). In this context, the digital photography method is both practical and reliably accurate, offering an optimal balance between precision and feasibility, which supports its applicability in real-world hospital settings.
From a sustainability and resource-efficiency perspective, routine monitoring of plate waste enables hospitals to identify high-waste meals and food components, thereby facilitating menu optimisation and portion size adjustment that may reduce unnecessary food production and disposal costs (Cook et al., 2022; Ni et al., 2026). Our findings and methodology provide a practical foundation for future studies integrating solid-food plate waste data with environmental impact assessments (e.g. carbon footprint) and cost analyses. This is particularly important given the growing emphasis at policy and institutional levels on the provision of sustainable healthcare services, alongside the need for scalable tools to support evidence-based food service management (Chatzipavlou et al., 2024; Cook et al., 2022).
4.2 Limitations
The digital photography method offers advantages in convenience, time efficiency, and reduced resource demands; its reliability depends on assumptions that may not hold in dynamic hospital foodservice settings, particularly consistent portion standardisation and minimal plating variability. Despite a quality assurance protocol to assess compliance with menu standards, deviations in serving practices may introduce error in baseline weights used for visual estimation. Variability in server technique, food density, ingredient substitutions, and presentation style may introduce systematic and random error, affecting accuracy and reproducibility. Without mechanisms such as inter-assessor calibration, reference libraries or periodic validation through direct weighing, data integrity may be compromised, particularly where nutritional adequacy or waste quantification are sensitive to portion volume. However, weighing each food component before and after serving is logistically challenging and disrupts workflow. A pragmatic alternative is periodic validation of portion sizes against hospital-standard serves (e.g. Menu and Nutrition Standards for Public Health Facilities in South Australia) via targeted direct weighing.
Additionally, liquid food products, such as soups and beverages, were excluded from photographic estimation due to practical constraints associated with hospital meal presentation. These liquid food products are routinely served in non-transparent cups or containers, which limits the ability to visually assess remaining volume from overhead tray photographs. While the digital photography method could be adapted for liquid foods under controlled presentation conditions, such adaptions were not feasible within a high-throughput hospital kitchen without disrupting standard foodservice workflows. In contrast, direct weighing is generally more efficient and operationally suitable for quantifying liquid waste, as it requires minimal handling and no modification to usual service practices. Combining photographic estimation for solid foods with direct weighing for liquids or side dishes provides a practical approach that balances measurement accuracy with operational feasibility in hospital foodservice settings. Within this scope, the present study therefore focuses on validating the digital photography method for estimating aggregate solid-food plate waste as commonly encountered in hospitals, while acknowledging that further validation is required for liquid-food products.
Furthermore, sauces were excluded from photographic estimations at lunch and dinner due to mixing with other foods, limiting visual assessment. Direct weighing captured sauce quantities, likely contributing to discrepancies between methods. Finally, this study also did not validate digital photography for estimating individual food items, reflecting practical constraints in isolating components, particularly with mixed dishes, overlapping items or returned plates where consumption intermixes components (Williams and Walton, 2011).
5. Conclusions
This study evaluated the accuracy of the digital photography method for measuring aggregate solid-food plate waste in an acute care hospital across breakfast, lunch and dinner. Plate waste estimates derived from the digital photography method showed good agreement with those obtained using direct weighing, supporting its validity for assessing aggregate solid-food plate waste. The accuracy of the digital photography method varied by mealtime, with higher accuracy observed for breakfast compared with lunch and dinner. These findings demonstrate that the digital photography method is a valid method for estimating solid-food plate waste in the hospital setting, while highlighting the influence of meal type on estimation accuracy.
Future research could extend validation to liquid food components such as soups and beverages, and to a broader range of diet types, including texture-modified and culturally diverse meals. Further work could also examine the applicability of this method in other healthcare contexts, such as rehabilitation hospitals, long-term care and aged care facilities, as well as explore the integration of photographic assessment with hospital nutrition monitoring systems to support more comprehensive evaluation of patient dietary intake and foodservice efficiency.
Authorship contribution
Huize Ni: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Resources, Validation, Visualization, Writing – original draft, Writing – review and editing. Taylor Willmott: Conceptualization, Methodology, Resources, Supervision, Validation, Visualization, Writing – review and editing. Dianne McGrath: Conceptualization, Methodology, Resources, Supervision, Validation, Visualization, Writing – review and editing. Helen Morris: Methodology, Project administration, Resources, Supervision, Validation, Visualization, Writing – review and editing. Tina Bianco-Miotto: Methodology, Project administration, Resources, Supervision, Validation, Visualization, Writing – review and editing.
Ethics statement
Publication endorsement was granted by the Central Adelaide Local Health Network Human Research Ethics Committee on 05 August 2025 (Reference Number: 21718).
The work has been supported by the End Food Waste Cooperative Research Centre, whose activities are funded by the Australian Government's Cooperative Research Centre Program and the University of Adelaide.
The authors would like to sincerely thank Dr Shao Jia Zhou for her valuable guidance and support, and Kimberly Lush for her assistance with the assessment of the inter-rater reliability of the plate waste estimates. Thank you also to the hospital foodservice staff and dietitians, as well as dietetics and nutrition students for their support and assistance with the food waste data collection.
The supplementary material for this article can be found online.
